The Lazy AI Strategy: Stop Hiring Young People

If AI leads your company to stop hiring juniors, you are doing AI wrong. Research on GitHub Copilot found productivity gains were largest among less experienced developers, meaning AI narrows the gap between juniors and seniors rather than making juniors redundant. Cutting junior hiring also destroys the pipeline that creates the senior engineers companies will need tomorrow.

If AI Means You No Longer Hire Juniors, You Are Doing AI Wrong

I recently read TechCentral’s article, “South African tech juniors squeezed as AI reshapes hiring.” It describes a disturbing shift in the technology labor market. Junior developers are struggling to get into the industry, entry level salaries are under pressure, and companies increasingly believe they can replace the bottom of the engineering pyramid with AI while competing aggressively for experienced engineers at the top.

Read the TechCentral article: South African tech juniors squeezed as AI reshapes hiring

The numbers should make every CIO uncomfortable. TechCentral reports that 62% of junior developers surveyed by OfferZen feel underpaid, that average entry level fintech salaries in its sample fell from R37,748 per month in 2025 to R27,777 in 2026, and that 48% of technology leaders feel pressure to hire more senior engineers as teams become leaner. TechCentral itself notes that the OfferZen sample skews toward Cape Town and Johannesburg, male respondents and employees at large companies. The sample is not nationally representative, but the direction is disturbing. The article ultimately asks the obvious question: if nobody hires juniors today, who exactly is going to become the senior engineer we desperately need tomorrow?

I think there is an even bigger question. What will history say about our generation of technology leaders if our response to AI was to pull up the ladder behind us?

1. We inherited a responsibility, not just a workforce

For most of human history, one generation taught the next. A father might take his son into the family business. A mother might teach her daughter a trade. Craftsmen took apprentices, doctors trained interns, accountants trained clerks, and engineers took graduates under their wings. The details changed across cultures and generations, but the social contract was remarkably consistent: come in, watch me, learn, make mistakes somewhere safe, become useful, and eventually become better than me.

That was not charity. It was how civilization transferred capability between generations, and we seem to be in danger of forgetting that.

The young person arriving at your company at 22 is obviously not as productive as the engineer who has been building systems for twenty years, but that was never the point. The 22 year old is there partly because of what they can contribute today and partly because of what they can become over the next decade. Somehow AI has encouraged us to evaluate that person only on the first half of the equation, and that is extraordinarily short sighted.

2. AI did not abolish apprenticeships

There is a seductive spreadsheet argument developing inside companies. A senior engineer with Claude, Codex, Copilot or another capable coding agent can now produce far more software than before, so rather than employing five engineers at different levels of experience, perhaps we employ two very experienced engineers armed with AI. The spreadsheet looks magnificent, for a while.

The mistake is assuming that because AI has automated some of the work traditionally performed by junior engineers, it has also automated the process through which humans become senior engineers. It has not. AI can generate code, but it cannot give someone ten years of lived engineering experience overnight. It cannot manufacture the instinct that something feels wrong before the dashboard turns red, and it cannot magically install judgment about distributed failure modes, organizational tradeoffs, customer behavior, architecture, security and operations. Those things are learned, and somebody has to give young people somewhere to learn them.

The available research on AI coding tools actually points the other way from the “juniors are obsolete” narrative. A randomized controlled trial of GitHub Copilot across Microsoft, Accenture and a large manufacturing company, published in Management Science, found that access to the tool increased the number of tasks developers completed in a week by roughly 26%, and the researchers report that this effect was larger for shorter tenure, less experienced developers. An earlier, more tightly controlled experiment by the same research team, cited in that working paper draft, assigned developers a defined task, implementing an HTTP server in JavaScript, and found that those with Copilot access completed it 55.8% faster than a control group, with the productivity gain again concentrated among less experienced developers. Research on GitHub Copilot found productivity gains were largest among less experienced developers, suggesting that AI can narrow some of the productivity gap between juniors and seniors rather than simply making juniors redundant.

Not every AI coding study finds a productivity gain, and a fair reading of the evidence has to include the ones that don’t. METR, an independent research organization, ran a randomized controlled trial in which experienced open source developers worked on mature repositories they knew well, and found that allowing them to use early 2025 AI tools made them 19% slower, not faster, even though the same developers believed AI had sped them up by 20%. METR has since said that newer tools probably close some of that gap, but a follow up study found the updated estimate too unreliable to publish, partly because developers who disliked working without AI increasingly dropped out of the control group. The honest summary of the literature is that results depend heavily on the task, the tool and the developer’s familiarity with the codebase, which makes the simple claim that “AI plus seniors beats juniors” harder to defend, not easier.

TechCentral makes precisely this longer term point: companies are increasingly demanding seniors while simultaneously shrinking the pipeline that creates seniors. That is not a talent strategy. It is eating tomorrow’s seed corn because this quarter’s food bill looks slightly lower.

3. The social consequences are much bigger than technology

There is another dimension that CIOs need to think about, because our responsibility does not end at the company firewall. What happens if an entire generation of young people discovers that the first rung of the economic ladder has disappeared?

This is no longer hypothetical. Stanford’s Digital Economy Lab, using ADP payroll data covering millions of American workers, found in its Canaries in the Coal Mine study that employment among 22 to 25 year olds in highly AI exposed occupations was 19% below where it would have been had it kept pace with less exposed occupations, up from a 13% gap when the researchers first documented it in August 2025. Experienced workers in the same occupations showed no comparable gap. More importantly, the researchers report that the effect was being driven primarily by reduced hiring of young people, not by companies firing existing employees. The first rung is already beginning to disappear.

Indeed’s Hiring Lab data tells a similar story from a different angle. As of May 2026, Indeed reports that entry level job postings in the US had declined 7.5% year on year, while postings for senior level roles rose 14.7% over the same period, with software development showing the widest seniority split of any sector they track. None of this proves that AI alone is responsible for the shift, since other economic factors are moving at the same time, but the direction is consistent across independent data sources and it matches exactly what TechCentral describes happening in South Africa’s fintech sector.

We tell children to study mathematics, learn science, learn programming, go to university, work hard, build skills and prepare themselves for the future. Then they arrive at the door and we tell them that AI can do the junior work and they should come back when they have ten years of experience. Where exactly are they supposed to acquire those ten years? Multiply that across software development, accounting, law, marketing, consulting, design, analytics and every other knowledge profession AI begins to reshape.

We could create something profoundly dangerous: a generation that is educated enough to understand the economy around them, ambitious enough to want to participate in it, technologically connected enough to see what everyone else has, but structurally prevented from getting onto the ladder themselves. That is not merely an employment problem, it becomes a social problem. Economically excluding millions of young people eventually has consequences for everyone, because economic participation creates dignity, independence, skills, families, tax revenue, consumers and social stability. Remove the route into productive work and the cost does not disappear, it simply reappears somewhere else in society.

CIOs therefore need to stop viewing graduate employment as somebody else’s problem. We are building the labor market our children will inherit.

4. The business case against juniors is also wrong

To be clear, AI may genuinely mean that some organizations need fewer software engineers for a given amount of output. I am not arguing that technology leaders should preserve yesterday’s headcount ratios forever. But “we need fewer engineers” and “we no longer need an entry level” are completely different conclusions. A healthy profession can shrink, grow or become dramatically more productive while still maintaining the pipeline through which inexperienced people become experienced ones.

There is something else bothering me about the argument. I do not actually accept that juniors have become economically unviable. I think many companies have simply failed to redesign themselves for a world in which juniors have AI.

A talented young engineer equipped with a frontier coding model has access to an extraordinary capability multiplier. They can understand unfamiliar code faster, generate tests, explore frameworks, explain errors, translate between languages, interrogate documentation and produce working software at a speed that would have been unimaginable for a graduate ten years ago. So why would we conclude that this makes the graduate less valuable? It should make them more valuable.

The problem is risk. A junior engineer with powerful AI can now make mistakes at extraordinary speed. They can generate a bad SQL query, create insecure code, misunderstand an architectural pattern or confidently deploy something they do not fully understand. But that is not an argument for removing juniors. It is an argument for fixing your engineering environment.

5. Build a company that is safe for young engineers

This is where CIOs should be concentrating their energy. If inexperienced engineers can destroy production, the problem is not inexperienced engineers, it is that your architecture is broken.

A well engineered technology organization should make it difficult for anyone, junior or senior, human or AI, to cause catastrophic damage. That begins with the software delivery pipeline. A junior engineer should be able to write code with AI, commit it and receive immediate feedback from automated tests, static analysis, security scanning, dependency checks, architecture rules and quality gates. Bad code should collide with guardrails long before it gets anywhere near a customer.

Your pipelines should test things humans routinely forget: concurrency, race conditions, failure paths, authentication, authorization, dependency vulnerabilities, performance regressions, schema compatibility, API contracts and infrastructure policies. The junior learns because the platform teaches them, and that is infinitely more scalable than depending entirely on a senior engineer staring over somebody’s shoulder.

6. Make AI infrastructure safe as well

The same principle applies when young engineers begin using AI agents. Do not give an AI coding agent unrestricted production credentials and then complain that juniors cannot safely use AI. Build the controls instead.

An AI agent exploring production data should ordinarily be routed towards read replicas, not primary transactional databases, and you should put a controlled proxy in front of those databases: inspect queries before execution, estimate their cost, enforce statement timeouts, limit concurrency, prevent destructive SQL, restrict DDL and DML, cap the number of rows that can be returned, and kill obviously pathological queries before they turn a production database into an expensive space heater.

This is not a theoretical idea. We have taken this approach ourselves. Capitec’s own pg proxy places a controlled layer between AI agents and PostgreSQL, analysing queries before execution and allowing dangerous or expensive queries to be blocked or shunted away from primary databases. Identity, data masking and observability live in the platform rather than relying on every developer to remember every rule. The principle matters more than the implementation: move safety out of people’s heads and into the engineering environment.

The same approach applies to cloud infrastructure more broadly. Use narrowly scoped IAM permissions, separate development and production accounts, apply policy as code, require approval for dangerous actions, detect anomalous infrastructure changes, and use ephemeral environments where engineers can experiment without gambling the company.

Then give young people room to explore. That is the important part. Guardrails should create freedom, not bureaucracy. The safest motorway is not one where nobody is allowed to drive, it is one deliberately engineered so that millions of imperfect humans can travel quickly without every small mistake becoming fatal. Our technology platforms should work the same way.

7. Turn your engineering platform into an apprenticeship system

Imagine the alternative to simply cutting graduate hiring. A junior joins your company. On day one they receive a development environment, an AI coding assistant, access to your engineering standards and a curated set of internal knowledge. They pick up a small piece of work.

AI helps them understand the repository and explains unfamiliar patterns, and it helps generate the implementation. Their local tools catch obvious mistakes. The pull request runs hundreds or thousands of automated checks, security tooling catches unsafe dependencies, and architecture tests detect prohibited coupling while integration tests expose incorrect assumptions. A senior engineer then reviews what actually requires human judgment, not whether somebody forgot a semicolon or used the wrong method name. The senior spends their scarce time teaching architecture, judgment, tradeoffs and context.

Suddenly the economics change. You are no longer paying an expensive senior engineer to supervise every keystroke of a graduate. You are using automation and AI to handle the mechanical parts of apprenticeship while preserving human mentorship for the things that matter. That can make juniors more economically attractive than they were before AI, not less.

8. Juniors may actually understand parts of the new world better than us

There is also some generational arrogance hiding inside the current debate. We assume experience automatically makes us better equipped for the AI era, and it doesn’t. A 22 year old entering technology today may have been using generative AI throughout university, and they may think about software creation completely differently from somebody who learned development twenty years ago. They might not know what we know, but we might not know what they know either, and that is precisely why companies need both.

Experienced engineers bring context, pattern recognition, judgment and scars. Younger engineers bring fewer assumptions, new behaviors, new tools and a willingness to approach problems without decades of organizational sediment attached to them. Combine those things and you have something powerful. Remove either side and you lose something important.

9. The lazy strategy is to remove the young

There will be companies that take the easy path. They will reduce graduate programs, stop hiring juniors, use AI to increase the output of their senior engineers, celebrate improved revenue per employee, and present beautiful productivity graphs to the board.

There is also a free rider problem hiding inside this strategy. For an individual company, it can look entirely rational to stop developing juniors and simply hire experienced people that somebody else trained. But if every company makes the same locally rational decision, there is nobody left doing the training. The industry keeps consuming a talent pool it has collectively stopped replenishing, which is exactly why competent, well intentioned executives can walk into this problem without ever making an obviously bad individual decision.

For several years, they may look brilliant. Then their senior engineers will leave, retire or become unaffordable, their institutional knowledge will thin out, and their succession pipelines will be empty. Suddenly they will discover that every company in the market is chasing the same small population of experienced engineers, because none of them bothered to create the next generation. The irony will be delicious, although the consequences will not be. Today’s supposed efficiency becomes tomorrow’s skills shortage.

10. Spineless leadership and the tyranny of the trailing metric

Somewhere along the way, leadership got quietly replaced by a spreadsheet. A CIO who watches revenue per employee climb because the graduate program was cancelled will show that graph to the board with visible pride, and the board will nod, because the number moved in the right direction this quarter. Nobody in the room asks what the graph looks like in five years, because nobody in the room is being measured on five years. They are measured on the next one.

That is not leadership, it is managed cowardice dressed up in the language of discipline and rigor. A spineless leader hides behind the metric precisely because the metric cannot be blamed later. If the senior engineers eventually leave and the pipeline turns out to be empty, the decision that caused it was made by “the numbers,” not by a person who could have chosen differently. Retrospective financial metrics are comfortable for exactly this reason: they describe what already happened rather than what is coming, and they let a leader claim to be rigorous and data driven while quietly avoiding the slower, harder, less measurable judgment calls that stewardship actually requires. When was leadership handed over to the trailing twelve months and the quarterly board pack? It did not happen overnight, but decades of executive pay tied to share price and quarterly earnings guidance have trained a generation of leaders to optimize for the number that gets read out on the call, not the one that will matter in a decade.

The frustrating part is that even on the numbers, the short term instinct is wrong. McKinsey Global Institute built a Corporate Horizon Index scoring hundreds of large listed companies on how genuinely long term their behavior was, based on patterns of investment, earnings quality and earnings management rather than what executives said in interviews. Comparing the two groups from 2001 to 2014, McKinsey found that the genuinely long term companies grew cumulative revenue 47% more and cumulative earnings 36% more than their short term peers, invested roughly half as much again in R&D, and, over the period to 2015, added close to 12,000 more jobs on average than companies optimizing for the short term. The lazy AI strategy of hollowing out the graduate pipeline to flatter this quarter’s headcount efficiency slots neatly into the pattern McKinsey describes: a decision that looks disciplined on a trailing metric and quietly destroys value on any horizon longer than a board cycle.

A lobotomised leader cannot be blamed for following the metric. A leader is supposed to look past it. That is the entire point of putting a human being with judgment in the role instead of an optimization algorithm, and it is precisely the judgment that a graduate pipeline exists to build in the next generation of leaders. If we let the trailing KPI make the call on juniors, we are not just failing this year’s graduates. We are training their eventual replacements to make the same cowardly call when it is their turn.

11. What kind of world are these leaders leaving behind

That is the question worth sitting with, because it is bigger than any single company’s headcount efficiency. If the trailing metric wins and the graduate pipeline keeps closing, the world these leaders leave behind is not an abstraction. It has a number attached to it, and in South Africa that number is already frightening.

Statistics South Africa’s own labour force data for the first quarter of 2026 put the NEET rate, the share of young people not in employment, education or training, at 37.6% for 15 to 24 year olds and 45.6% for the broader 15 to 34 year old group. That means close to half of the young people in this country are neither working nor studying nor being trained for work. This is not a distant risk that AI might someday create. It is the starting line AI-driven hiring caution is now being layered on top of, and every graduate program that gets quietly shelved to protect this quarter’s margin adds to it.

The academic literature on what happens next is not comforting. Researchers studying youth unemployment across East Africa and the wider region have documented a consistent link between large numbers of excluded young people and political instability, and similar patterns of frustration, social exclusion and reduced trust in institutions have been documented among unemployed graduates elsewhere. Nobody is suggesting a graduate hiring freeze at one bank causes a riot. But the mechanism is well understood: a generation that did everything it was told to do, studied, qualified, showed up, and still finds no door open to it, does not quietly disappear. It becomes disengaged, or it becomes angry, and either way the cost lands on the same society the company operates in, just later and paid by someone else.

That delay is precisely what makes the trailing metric so seductive and so dishonest. The CIO who cancels the graduate intake this year will not be in the room when the NEET numbers show up in reduced tax revenue, strained social services, or a generation with less trust in the institutions that failed to make room for them. By then the spreadsheet will show someone else’s name at the top. That is the world these leaders are leaving behind: not a crisis they caused on purpose, but one they were entirely capable of seeing coming and chose not to price in, because nothing on this year’s board pack asked them to.

12. CIOs will be judged for this

Leadership is not simply optimizing the current quarter, it is stewardship. We inherited companies, institutions, professions and knowledge from people who came before us, and most of us were once given an opportunity by somebody who knew considerably more than we did. Someone tolerated our questions, reviewed our terrible code, explained why the database had just fallen over, trusted us with something slightly bigger than we were ready for, and invested in us before the business case was completely obvious.

Eventually, we became the experienced people in the room. Now it is our turn. AI does not release us from that responsibility, and if anything, it increases it. We have been handed perhaps the most powerful productivity technology of our careers. We can use it to concentrate opportunity among those who already have experience, or we can use it to make knowledge, capability and opportunity accessible to a much larger generation. That is a leadership decision.

13. Build the on ramp

So my message to CIOs is simple. Do not remove the juniors, build a better on ramp. Make your engineering environment safe enough for inexperienced people to contribute. Automate testing and security, build strong CI/CD controls, separate production from experimentation, put guardrails around databases, constrain AI agents, and make observability excellent. Create clear progression frameworks, give juniors real work, pair them with experienced engineers, and allow them to make mistakes in places where mistakes are inexpensive. Then gradually increase the size of the blast radius they are trusted to manage.

That is how you create senior engineers. It is how we have always created senior engineers. AI simply gives us an opportunity to do it faster and better.

14. Don’t pull up the ladder

There is a version of the AI revolution in which a relatively small group of experienced, highly paid people becomes enormously productive while companies steadily close the doors through which the next generation would once have entered. That version might maximize a few corporate metrics, but it would be disastrous leadership.

The alternative is much more interesting. We use AI to give young people capabilities previous generations could never have imagined. We surround them with engineering systems that make experimentation safe. We combine their energy and curiosity with the judgment of experienced people. We shorten the journey from novice to expert without pretending the journey no longer matters.

Twenty years from now, when somebody asks what our generation of CIOs did when AI arrived, I hope the answer is not that they used it to stop hiring our children. I hope the answer is that they built the systems that gave an entire generation a way in.